Releases: lucidrains/se3-transformer-pytorch
Releases · lucidrains/se3-transformer-pytorch
0.0.22
add fourier encoding of distance as a feature
0.0.21
save extra projection out for 1 headed attention (plain non-multihead…
0.0.20
add backup loading from numpy
0.0.19
allow for token to attend to itself by default
0.0.18
allow gradients from basis vectors -> r_ij with a flag change, sugges…
0.0.17
add an extra norm at the end, like traditional transformers
0.0.16
add reversible networks
0.0.15
allow for float64 across whole model
0.0.14
turn off null key/values feature
0.0.12
fix device problem for mask